Professional Documents
Culture Documents
Study Quality Guide: Cochrane Consumers & Communication Review Group
Study Quality Guide: Cochrane Consumers & Communication Review Group
CONTENTS
This guide was prepared by Rebecca Ryan and edited by Dominique Broclain, Dell Horey, Sandy Oliver
and Megan Prictor. This paper was approved by Sophie Hill, Coordinating Editor of the Consumers and
1
Study Quality Guide May 2013
Communication Review Group on 5 March 2007. It was substantially revised in 2011 by Rebecca Ryan,
Sophie Hill, Megan Prictor and Joanne McKenzie. Last updated: 30 May 2013
People using this guide should check the website http://cccrg.cochrane.org/author-resources regularly to
ensure they have the latest version.
2
Study Quality Guide May 2013
2.0 INTRODUCTION
Structure
Section 1: What is risk of bias?
Section 2: Why assess risk of bias?
Section 3: Bias: what you should consider
Section 4: How to include risk of bias assessment in systematic reviews.
Section 5: Risk of Bias assessment: randomised controlled trials and non-
randomised controlled trials
Section 6: Quality assessment: qualitative studies.
Section 7: How to report risk of bias: tips for authors.
This document must be read in conjunction with the Cochrane Handbook for Systematic Reviews
of Interventions (Higgins 2011), which gives extensive guidance on the conduct of systematic
reviews. Referencing to appropriate sections of the Handbook is given throughout. Readers are
also referred to the Cochrane Collaboration Open Learning Material for Reviewers, Version 1.1,
November 2002 (Updated in 2009), which provides additional information on methodological
issues in conducting a systematic review. More recent online training materials are available for
Cochrane contributors are http://training.cochrane.org/
Additional sources of information for review authors, with brief descriptions of their key points,
are provided in Appendix A. Appendices to the accompanying Study Design Guide contain
glossaries of useful terms for study design and the characterisation of study design. These are
adapted primarily from the Cochrane Glossary, with additional material from other sources.
3
Study Quality Guide May 2013
Quality is itself a difficult concept to qualify or to quantify. It’s a complex idea that means
different things to different people1, and there are numerous tools in existence with which to
assess study quality. However, there is little evidence available to help guide the selection of
quality assessment tools to be used in systematic reviews2. Further, the reporting of quality in
published studies is often poor, which can increase the difficulty of assessing relevant
information.
Despite these difficulties, most definitions of quality or validity3 used in systematic reviews
involve some measure of the methodological strength of the relevant study, or how able it is,
through its design and its conduct, to prevent systematic errors, or bias.
The following is taken directly from the Cochrane Handbook for Systematic Reviews of
Interventions, Section 8.1.
8.1 Introduction
The extent to which a Cochrane review can draw conclusions about the effects of an
intervention depends on whether the data and results from the included studies are
valid. In particular, a meta-analysis of invalid studies may produce a misleading result,
yielding a narrow confidence interval around the wrong intervention effect estimate.
The evaluation of the validity of the included studies is therefore an essential
component of a Cochrane review, and should influence the analysis, interpretation and
conclusions of the review.
The validity of a study may be considered to have two dimensions. The first dimension
is whether the study is asking an appropriate research question. This is often described
as ‘external validity’, and its assessment depends on the purpose for which the study is
to be used. External validity is closely connected with the generalizability or applicability
of a study’s findings, and is addressed in Chapter 12.
The second dimension of a study’s validity relates to whether it answers its research
question ‘correctly’, that is, in a manner free from bias. This is often described as
‘internal validity’, and it is this aspect of validity that we address in this chapter. As
1
For example, it can mean the internal and/or external validity of a study, the clinical relevance of the research
question that is studied, the appropriateness of the data analysis and the presentation of the findings, or the ethical
implications of the interventions under evaluation.
2
There is limited empirical evidence on the relationships between specific aspects of study quality and the actual
findings of studies. There is some evidence that studies which lack allocation concealment and proper blinding may
overestimate the effects of interventions, but the effects of other elements of quality on outcomes has not yet been
established empirically.
3
Validity, in general terms, is the extent to which the result (of a measurement or of an intervention/ study) is likely
to be true and free from systematic errors.
4
Study Quality Guide May 2013
Variations in study quality can explain differences in the findings of studies that are included in
a systematic review. As a result, the quality of a study will reflect the strength of the evidence
that can be drawn from it. In other words, it reflects the confidence that we can have that the
results of a study reflect the ‘truth’. By extrapolation, the quality of studies included in a
systematic review will affect the confidence we can have in the results of the systematic
review. See resources listed in Appendix A of this guide for further information on quality and
validity.
Generally, assessment of study quality includes assessment of at least some elements of the
internal validity of the study4. This is the degree to which the design and the conduct of the
study avoid bias (Jadad 1998). Simply put, it is the degree to which we can have confidence
that the results of the study reflect what is ‘true.’ Higher quality studies are more likely to
produce results that are closer to the true result, as they are less prone to bias or distortions
from the true value. Evaluating the internal validity of a study therefore involves assessing the
measures that were used to try to prevent or minimise bias. These might include evaluation of
elements such as randomisation, allocation concealment and blinding.
The guidance in this document focuses mainly on the assessment of internal validity and
related issues for studies included in systematic reviews. Internal validity is the least context-
dependent aspect of study quality, and some elements of internal validity are empirically
related to study results. For example, trials with inadequate allocation concealment have been
shown to consistently overestimate the effects of interventions. Assessment of study
quality should therefore always include an assessment of internal validity.
Other aspects of study quality may also be considered and evaluated according to how
relevant they are to the scope and to the particular question(s) addressed by the review. For
instance, studies included in systematic reviews should also be evaluated in terms of external
validity. This is the extent to which the results of the study can be applied, or generalised,
to the population outside the study; in other words how the study’s results apply to the real
world.
4
Internal and external validity: Internal validity is the extent to which the study’s design, conduct, analysis and
presentation have minimised or avoided biased comparisons of the intervention under investigation. External validity is
the precision and extent to which it is possible to generalise the results of the study to other settings (ie. to real life
clinical situations) (Jadad 1998).
5
Study Quality Guide May 2013
That said, evaluating internal validity is necessary, but not sufficient, to comprehensively
evaluate a study included in a systematic review. The guide that will follow on the analysis and
interpretation of the results of studies included in a systematic review [still in development,
May 2011] deals in more detail with issues relating to the interpretation and external validity
of studies. The current document will focus primarily on aspects of quality relating to internal
validity. It will address the quality assessment of randomised controlled trials (RCTs) and of
studies with non-randomised design, as both are relevant to the scope of the Cochrane
Consumers and Communication Review Group and may be included in systematic reviews
coordinated by the Group.
Much of the research on quality assessment to date has focussed on the evaluation of RCTs. As
a result, this is where the most comprehensive advice is currently available. However, different
aspects of quality apply to studies of different designs, and recently there has been increasing
interest in methods of assessing and reporting quality for a range of different study types. This
document attempts to provide guidance to authors on quality assessment of both randomised
and non-randomised studies.
An important point to be aware of when conducting a systematic review is that the methods
you have employed throughout the review need to be transparently reported. You should
aim to produce a review that is reproducible by anyone who reads it. That is, anyone should
be able to conduct the review on the basis of what you have written and come to similar
conclusions. This is only possible if you explicitly report the details of each decision you make
throughout the review process.
All decisions that you make throughout the course of conducting and writing your review need
to be logical and justified. They must also be clearly and consistently reported. While these are
important issues throughout the entire review process, they may be especially important when
you start to consider how you will evaluate the quality of included studies (which criteria you
will evaluate studies against); how you will report study quality as a component of the
systematic review; and how you will incorporate quality assessment into interpreting the
results of the review.
For example, if your review includes only RCTs, how will you decide if a study is really an RCT?
For studies that report that they ‘randomly’ assigned participants but do not provide any
details about their randomisation method, will you decide to include or exclude them from your
review? Will you await response to author contact to confirm that they used a truly randomised
technique? What will you do if the authors are not contactable? Decisions like these that you
6
Study Quality Guide May 2013
make during the review process need to be explicitly reported, so that the logical processes
you have followed throughout the review are clear to readers.
There are many complex decisions to make when conducting a systematic review. These
decisions may be further complicated by variable quality of included studies, poor quality of
reporting, complex interventions, complex outcomes or means of outcome assessment, or by
other factors. Such factors may make the task of systematically reviewing the evidence on a
particular intervention an involved one. However, these factors are not in themselves
problematic, as long as you are transparent and systematic in the way that you report these
issues and the decisions that you reach as you progress through the review.
7
Study Quality Guide May 2013
Studies, even well-designed ones, are never perfect in their design or execution, and there are
many different ways in which they can be compromised or limited. These limitations (or
various biases, see the following Section 2.3) can affect the results of a study. Biased studies
are more likely to produce misleading results than those that are rigorously designed and
conducted.
A systematic review is only as good as the studies upon which it is built. Including biased
studies in a systematic review can therefore produce misleading results. Even if high quality
methods are followed for the conduct of the review itself, if studies with serious biases are
included and these are not adequately accounted for or acknowledged, poor quality evidence
will arise from the review.
Assessing the risk of bias of a study that might be included in a Cochrane review can be useful
in a number of ways.
Authors can set a minimum threshold in order for a study to be included in the review.
For example, review authors may decide to exclude all RCTs that do not have adequate
sequence generation (randomisation). Note that such decisions should be made and
reported in advance, or a priori.
A second use for risk of bias assessment can be to initially include studies at higher risk
of bias, but to successively remove them from the analysis. This can allow review
authors to determine how much, if at all, the result is affected. This approach is called a
sensitivity analysis as it tests how ‘sensitive’ the results are to the risk of bias of the
included studies.
Finally, risk of bias assessment of the included studies can be used to help guide and
structure the discussion and interpretation of the review’s results, to assist in
determining the strength of evidence that can be drawn from the review, and to guide
recommendations or implications and directions for future research.
8
Study Quality Guide May 2013
Variations in the validity of studies can explain differences in the results of the studies that are
included in a systematic review. More rigorous studies – or those at lower risk of bias - are
more likely to yield results that are closer to the ‘truth’. Analysis of results from studies of
variable validity can result in false conclusions being drawn about the effects of an
intervention5. Even if there is no variability in the results of the studies included in a
systematic review, it is still important to systematically evaluate, or critically appraise, all
of the included studies. In such an instance, while the results may all be consistent, (that is,
they report similar effects of the intervention on specific outcomes), the studies from which
they come may still have serious methodological flaws. This means that the conclusions drawn
from the review would be much weaker: you could have less confidence that they accurately
reflect the true effect of the intervention than if the studies had all been rigorously designed
and conducted and had yielded consistent results.
Therefore it is vital to be able to recognise and systematically evaluate the likely contribution
of different sources of bias in included studies. It is essential to assess whether and how
these may have affected the results in order to determine how confident you can be in the
study’s results, and that they reflect reality rather than the influence of other factors.
See Systematic Reviews; CRD’s guidance for undertaking reviews in health care; 1.3.4 Quality
assessment, available at http://www.york.ac.uk/inst/crd/pdf/Systematic_Reviews.pdf
While design and quality are related characteristics of any study, they actually represent
different components of a study. Design refers to how the study has been set up, or its
framework. Quality, in comparison, is a complex concept that reflects how well the study was
designed and how well it was executed. Hence, while the design of a study affects its quality
(in part, how able it is to prevent bias), both quality and design determine the strength of the
evidence that can be drawn from the results of any particular study.
An important point to realise is that not all study designs are equal when it comes to
answering different types of research question. The ranking or hierarchy of different study
designs depends on the question that is being asked (see the Group’s Study Design Guide
for more information; see also the Cochrane Handbook for Systematic Reviews of
Interventions, Section 5.5; and the CRD’s guidance for undertaking reviews in health care,
Section 1.2.2.3). When considering studies of effectiveness (that is, the effectiveness of
5
For example, if the study overestimates the effects of an intervention, the study may make a ‘false positive’
conclusion (ie. wrongly conclude that the intervention works). If, in comparison, the study underestimates the effects
of an intervention, it may make a ‘false negative’ conclusion (ie. wrongly conclude that there is no effect of the
intervention).
9
Study Quality Guide May 2013
therapy or some other type of intervention), the questions tend to focus on comparisons, or
how an intervention compares with no intervention or with alternative intervention(s). To
answer questions about the effects of interventions– which is what most Cochrane reviews
address - comparative studies that minimise bias will be the highest in the hierarchy, or the
most suitable types of studies to investigate these questions.
It is for these reasons that RCTs are regarded so highly. RCTs are considered to be the ‘gold
standard’ for addressing questions of effectiveness, as they are designed to minimise bias. For
example, randomly assigning participants to treatment groups reduces the likelihood that the
groups will differ on important baseline characteristics. This includes those characteristics of
which the investigator may not be aware. By using chance it is likely that the two groups will
be equivalent on important characteristics. Although it is possible to control for known
confounders using other study designs,6 randomisation is the only was to control for
confounders that are not known. This means that selection bias is likely to be minimised in
RCTs; whereas the means of allocating participants in other study designs may not be as ‘fair,’
and groups of participants may differ on important characteristics at baseline.
Many different ways of classifying or ranking studies have been developed7. Although these
schemes differ, broad similarities also exist. Below (Figure 1) is a general classification scheme
or hierarchy of studies in terms of the suitability of their design to answer questions of
effectiveness. Such hierarchies are based on the susceptibility of different study designs to
bias.
RCTs
Non-randomised studies, that is:
Quasi-randomised controlled trials
Controlled before-and-after studies (CBAs)
Interrupted time series studies (ITSs)
Controlled observational studies
Cohort studies
Case-control studies
Case series
6
For example, by ‘matching’ participants in the different study groups on important characteristics.
7
See Systematic Reviews; CRD’s guidance for undertaking reviews in health care; 1.3.4 Quality assessment (available
at http://www.york.ac.uk/inst/crd/pdf/Systematic_Reviews.pdf). The exact schema adopted is not essential, but it is
important to have some idea of the general level of evidence that different types of study may represent.
10
Study Quality Guide May 2013
Many Cochrane systematic reviews specifically include RCTs only, as they are considered the
strongest type of evidence to answer effectiveness questions. Reviews that fall within the scope
of the Consumers and Communication Review Group, however, include many interventions for
which an RCT may not exist. Sometimes interventions have not or cannot be subjected to an
RCT, or even to a quasi-RCT, for practical or ethical reasons. Further, many kinds of research
question may benefit from the inclusion of non-randomised studies, such as public health and
health promotion interventions.
We have therefore decided to follow the advice developed by the Cochrane Effective Practice
and Organisation of Care (EPOC) Review Group (see also http://epoc.cochrane.org/epoc-
resources). This advice allows review authors to decide whether to include a limited range of
experimental study designs other than RCTs in their analysis and review.
These guidelines state that the following study designs are eligible for consideration for inclusion
in systematic reviews of complex interventions:
RCTs (including cluster RCTs)
Non-randomised studies
Quasi-randomised controlled trials
Controlled before-and-after studies (CBAs)
Interrupted time series (ITS)
It should be noted that any hierarchy of study designs should be treated with flexibility and
used as a guide only. For example, RCTs should rank at the top of such a hierarchy only when
they are well-conducted.8 This is why it is necessary to assess the risk of bias of the studies
included in a review – even if it only includes RCTs – because not all studies of the same basic
design will be equally well conducted, and this may affect the results and conclusions of the
review.
The nature and type of the review question, as well as pragmatic considerations such as author
resources, should determine the choice of studies to include in a review. In terms of providing
answers to a review question, the suitability of different studies should be obvious if the
question is formulated clearly and explicitly. A review should be based on the best quality
evidence that is available. If authors decide to include studies other than RCTs in their review,
that their reasons for doing so must be clearly and explicitly stated.
Think about the type of question you want to answer with your systematic review. What
types of study design would be most appropriate to answer this question?
8
If there are serious flaws or limitations in the design or conduct of an RCT, it should be considered as being of lower
quality than one that is performed well.
11
Study Quality Guide May 2013
On the basis of the above, what studies will you include in your review? Will you include
studies other than RCTs? If you decide that it is necessary to do so in order to answer
your review question, you will need to explicitly justify your decision in the review.
This must be stated clearly at the protocol stage of your review, together with
a rationale for your decision.
As study designs can be fundamentally different in their structure and conduct, they
also need to be differentially assessed for quality. That is, different elements of quality,
or quality criteria, need to be considered for different study designs. Specific quality
items to be assessed for each different type of study design to be included in
the review must be clearly and systematically described at the protocol stage.
These items must be described clearly and separately for each of the different study
designs to be included in the review.
For further information, see Systematic Reviews; CRD’s guidance for undertaking reviews in
health care; 1.3.4 Quality assessment
(available at http://www.york.ac.uk/inst/crd/pdf/Systematic_Reviews.pdf )
A major problem when assessing study quality is that the reporting of studies is often poor.
Authors often neglect to provide adequate details of the methods they used to reduce bias.
Even if the methods are reported, many authors simply report such measures as being ‘done,’
rather than providing detail that might allow an independent evaluation of the adequacy of the
approach they used. This often makes the systematic assessment of study quality quite
difficult. In cases where there is inadequate reporting, a review author may decide to assume
that adequate measures were taken, or alternatively, that they were not taken9. Either
decision may, however, be incorrect and may lead to an inaccurate representation of the study
when included in the systematic review.
Given that many journals and other publications impose strict word limits on study reports, it
is perhaps not surprising that authors often fail to elaborate on their methods. Standards for
the reporting of trials have now been developed, and will help to standardise and improve the
9
That is, as a review author, you could decide to consistently rate those instances where there is insufficient detail
available in a consistent manner as either done or not done. In either case you should report this decision in the text
of your review.
12
Study Quality Guide May 2013
reporting of future studies10. However, those studies published previously remain likely to pose
significant problems for review authors, when insufficient detail is reported to fully assess
study quality11. This is often true in the case of Cochrane reviews, where the inclusion of
studies is typically not restricted by the year of publication and it is common for a review to
include studies dating back several decades.
Rather than assuming that something was not done if it was not reported, a more
thorough approach should be taken. Review authors should contact the study authors in
the first instance to request further information. If this is not possible or not successful12, the
reporting of quality should reflect this underlying uncertainty. The Cochrane Risk of Bias tool
(covered in more detail in following sections) encourages this type of transparent reporting
and provides a format for providing this type of information within a review. For example, if
details about a particular issue are not provided in a trial report, this can be reported as ‘not
described’.
Building in this capacity for reporting uncertainty may allow more accurate reporting on the
status of the study than simply assuming that critical elements of the study were done or not
done. This may, in turn, be more meaningful when it comes time to interpret the study’s
results and to incorporate them into the review’s overall findings, and is also informative to
readers of the review.
Variations in study quality (validity) can explain differences in the results of studies
included in a systematic review. More rigorous or higher quality studies are more likely
to closely estimate the truth. Study quality should therefore be assessed
systematically, transparently and in an explicit manner.
Poor reporting of elements of a study that relate to quality is not the same as poor
quality of the study itself.
The Consumers and Communication Review Group recommends that if not enough
detail is provided to assess quality criteria for included studies:
o Review authors should contact the study authors to see if more information about
the design and/or conduct of the study can be obtained (for sample letters contact
the Managing Editor).
10
For example, the Consolidated Standards of Reporting Trials (CONSORT) statement was developed to improve the
reporting of RCTs (see www.consort-statement.org); while the Transparent Reporting of Evaluations with Non-
randomised Designs (TREND) statement provides guidelines for the standardisation of reporting for non-randomised
studies (see www.cdc.gov/trendstatement/). See Appendix A of this guide for further information.
11
While the reporting of study quality is always improving, older studies may be more limited in their reporting. As a
review author, you need to be consistent and clear about how you intend to deal with studies where a lack of
information or detail is problematic in their evaluation.
12
For example, for studies conducted a long time ago (10 or 20 years), even if study authors can be contacted they
may no longer have access to the data or the specific details of the conduct of the study.
13
Study Quality Guide May 2013
o Where further information is not available, it is possible to report this directly using
the Risk of Bias tool within reviews.
Bias is any systematic error or factor, whether it is recognised or not, that distorts the results
or inferences of a study. It can occur because a study design is poor, or because a study is
poorly conducted. As mentioned above, although some study designs are better than others at
avoiding or minimising bias, you cannot assume that all studies of a particular design will have
been conducted equally well, or as well as they might have been. This is why it is necessary to
assess the risk of bias of all studies included in a review: to be able to tell how well likely it is
that there are systematic errors that might affect the results of the studies and the review.
Bias can arise from many different sources. There are five major sources of bias that can arise
in studies examining the effects of healthcare interventions:
Selection bias
Performance bias
Attrition bias
Detection bias
Reporting bias
The table below is taken directly from The Cochrane Handbook for Systematic Reviews of
Interventions, Section 8.4.
14
Study Quality Guide May 2013
Selection bias occurs at the point of allocating participants to groups at the beginning of a trial,
and results from systematic differences in the characteristics of the participants in each study
group. These differences can relate to prognostic factors, or can be due to the differential
responsiveness of participants to the intervention.
13
Blinding prevents participants, providers and/or outcome assessors from knowing which participants are receiving
the intervention(s) in the study.
15
Study Quality Guide May 2013
assigned to the different groups before it actually occurs. In comparison, blinding (whether
of participants, providers or outcome assessors) refers to measures that are taken after
people have been assigned to groups. This means that no-one knows which participant
belongs to the different groups throughout the course of the study. An important difference
between the two is that allocation concealment can always be implemented, whereas
blinding cannot always be successfully achieved.
Allocation concealment has been shown to impact importantly on the results of trials.
Several studies have demonstrated that when allocation concealment is inadequate or
unclear, the treatment effect is overestimated.
Allocation concealment can be achieved in different ways, but different approaches can
have varying levels of success; that is, they are more or less likely to be interfered with or
subverted.
RANDOMISATION
‘Truly’ random methods of generating the randomisation sequence (ie. methods that produce
a non-predictable assignment pattern):
o Computer-generated random numbers
o Random number tables
o Coin toss (for a trial with two groups) or die toss (where there are
more than two groups)
Note: Trials employing a truly random sequence generation method are designated as RCTs.
16
Study Quality Guide May 2013
Central computer randomisation and allocation (for example, allocation by a central office
that is unaware of the participants’ characteristics);
On-site computer from which assignment can only be determined after entering the
patient’s data;
Pre-numbered or coded identical containers administered serially to participants;
Serially (sequentially) numbered, opaque sealed envelopes.
There are also other approaches, similar to those above, that if administered by a different
person to the one who generated the allocation scheme can also be considered adequate.
Certain other approaches might also be new or innovative and not fit with those described
above but can still provide adequate allocation concealment.
Any ‘open methods’ (ie. transparent before allocation) eg lists on noticeboards, open lists,
open envelopes;
Non-opaque envelopes;
Odd or even date or medical record number;
Dates of birth or days of week;
Alternation.
In cases where studies do not report any approach to concealing the allocation sequence, it
should be assumed that the adequacy of approaches is unclear. Examples of this might
include cases where studies state that a list or table or random numbers was used but do not
provide further details; or stating that sealed envelopes were used .
Consider:
Was there any way that those conducting the study could have known what the random
allocation sequence was? Could allocation to groups be manipulated?14
Were the study groups comparable at the start of the study, just after randomisation?
This bias arises from systematic differences in the way that care is provided, or from exposure
to factors other than the intervention that is being studied.
Performance bias occurs during the treatment and/or delivery of the intervention(s).
14
If the answer to this questions is ‘yes,’ the method of allocation concealment can be considered inadequate.
17
Study Quality Guide May 2013
It arises as the result of differences in the way that the intervention is delivered to the
different study groups; that is, not only does the intervention differ between groups, the
method of delivering it also differs. The impact of the intervention alone therefore cannot
be assessed.
Consider:
Were there any other factors that might have affected the study’s results?
Of those involved in the study, were the following blinded:
o Participants?
o Provider/ caregivers?
We suggest that you avoid the use of terms like ‘double blind’ etc. unless you clearly and
explicitly define them in your review (eg. ‘we will use the term double blind to mean…’).
Instead we suggest that you specifically report on each component of blinding separately
(eg. whether participants were blind and how adequate the approach to blinding was;
whether providers were blinded adequately, and so on). This makes it very clear who was
blinded throughout the study; allows comment on the adequacy of each possible
component of blinding; and allows the potential bias introduced by inadequate blinding to
be identified clearly both by you the review author and the readers of your review.
Blinding of outcome assessors is addressed under Detection Bias (at 2.3.3) below.
Blinding of data analysts and manuscript writers is not explicitly addressed in the Cochrane
Collaboration’s Risk of Bias tool.
15
There is very little consensus regarding what actually constitutes single/ double/ triple blind, and although these
terms are commonly used there is little consistency in the way that they are used. See Appendix A of this guide fore
more on blinding and interpretations of blinding.
18
Study Quality Guide May 2013
This bias arises due to differences in the way that outcomes are assessed. Detection bias
occurs at the point of follow-up (that is, at the time of outcome assessment). The Cochrane
Handbook says:
All outcome assessments can be influenced by lack of blinding, although there are
particular risks of bias with more subjective outcomes (e.g. pain or number of days with
a common cold). It is therefore important to consider how subjective or objective an
outcome is when considering blinding. The importance of blinding and whether blinding
is possible may differ across outcomes within a study…
Blinding of outcome assessment can be impossible (e.g. when patients have received
major surgery). However, this does not mean that potential biases can be ignored, and
review authors should still assess the risk of bias due to lack of blinding of outcome
assessment for all studies in their review.
Consider:
Were outcome assessors blind to the treatment allocation?
This bias arises due to systematic differences between study groups in the withdrawals or
exclusion of people entered into the study. That is, it is due to systematic differences between
study groups in the numbers of participants who are ‘lost’ during the study (ie. it occurs after
the allocation and inclusion of participants in the study). Attrition bias occurs over the
duration of the trial.
16
Note that allocation concealment helps to prevent selection bias and protects the randomisation sequence before
and until the interventions are given to the study participants. In contrast, blinding helps to prevent detection bias, by
protecting the randomisation sequence after allocation.
17
Attrition bias is also known as exclusion bias, as it may result from the post-randomisation exclusion of
participants from the study. See Appendix A of this guide for more information on attrition bias.
19
Study Quality Guide May 2013
For example, in a study examining the effects of a particular drug, more people receiving the
active drug may leave a study due to side effects than those assigned to the control group.
This could lead to an overestimation of the effectiveness of the intervention, as only those
people who tolerated the intervention well remained in the study and were assessed. Attrition
bias could also lead to an underestimation of the adverse effects or harms associated with the
intervention, if those with the most severe side effects left the study and were not measured
and included in analysis.
In the case of RCTs, random assignment is not itself sufficient to make inferences infallible.
The validity of inferences depends on the assumptions that random assignment produced
comparable groups at baseline, and that systematic attrition over the study period did not
occur. Studies often fail to report how many participants who were enrolled in the study were
followed throughout it, which makes sample attrition difficult to assess. Even if not reported
however, it is important to note the attrition of the samples as a possible source of bias; and
to interpret the results of studies with large rates of attrition with some caution.18
ITT analysis is a way of preserving the equivalence of the groups established at the start of
the study (baseline) by the process of random allocation. If the study participants are
randomly assigned to groups and there are adequate methods of allocation concealment,
the groups should (theoretically) be similar at the start of the study and should be analysed
on this basis. In order to prevent attrition bias, the groups also need to be similar at the
18
The way that losses of participants throughout the study are handled has much potential to introduce bias. However,
reported attrition following allocation has not been found to be clearly related to bias. It is not yet clear exactly what
the relationship between attrition and bias are, so review authors should be cautious when dealing with this aspect of
study execution.
20
Study Quality Guide May 2013
end of the trial (Heritier et al 2003). The alternative approach, to analyse participants
according to whether the participants did or did not receive the intervention, can markedly
change the characteristics of the different study groups, and so make the process of
random allocation ineffective.
Often studies report outcomes only for those participants that were followed up or
completed the study; or they do not provide enough information to determine whether ITT
analysis was used. Even if ITT analysis is not used, it is important that the participants lost
from the study (after they were enrolled in the study) are clearly accounted for. The way in
which they were dealt with in the study’s analysis must also be clearly and completely
described. Many studies fail to report any details on these issues.
Often the numbers of participants involved throughout studies (participant flow) are poorly
reported, although this is improving in more recent trial reports. Review authors should
contact study authors to request further information about participant flow if it is not clear
from the study report.
Note that even where participant flow is reported, participant numbers are often implied
rather than explicitly stated or reported at each stage of the study. Authors of such studies
should be contacted to provide further information to clarify participant flow. If information
is not available, the potential bias introduced by participant attrition from the study should
be noted in the review.
Consider:
Was follow-up adequate?
o How many participants completed the study for each group and for each outcome?
o How many participants did not complete the study for each group and for each
outcome?
Were the study groups comparable at the end of the study?
Were those followed up comparable to those who were lost from the study?
Was ITT analysis used?
If ITT analysis was not used, how were those who dropped out (or were excluded post
randomisation) treated in the analyses?
21
Study Quality Guide May 2013
See Cochrane Handbook for Systematic Reviews of Interventions, Section 8.14 and
Section 10
See also Egger et al (2003) and Song et al (2004) for more information
This bias arises because of the existence of systematic differences between findings that are
reported and those that are not. This can happen in a number of ways.
Within studies
Selective outcome reporting: it is possible for only some outcomes to be included in the
trial report. This means that some of the outcomes have been omitted from the report. For
example, a study might fail to report on all primary outcomes that were pre-specified; may not
report on a key outcome that would usually be reported for that type of study; may report
data incompletely for one or more outcomes; or may otherwise fail to convince a reader that
all outcomes that were pre-specified have been fully reported. A study can also incompletely
report all data for a given outcome, for example, selectively reporting data only at some but
not all pre-specified time points. Selective outcomes reporting arising in any of these ways
may systematically distort results and introduce bias.
Across studies
Reporting biases: at the study level, it can also happen that the direction of the research
findings affects the dissemination of the research. The best-recognised type of reporting bias is
known as publication bias.
Publication bias arises due to the fact that studies with statistically significant ‘positive’
results are more likely to be published than those studies with results that are not. This means
that in the collected published research literature on a given topic there are less likely to be
studies reporting non-significant results. This introduces a systematic error into the collected
research evidence, for example, systematic reviews that do not include unpublished studies
may be more likely to overestimate the effects of the intervention.
There are also other sorts of reporting biases that can be introduced when conducting a
systematic review if the search strategy is not comprehensive19. Examples are shown in the
table below.
19
Publication and associated biases can be formally evaluated using various statistical techniques, including
funnel plots. See the Cochrane Handbook chapter 10.4 for details.
22
Study Quality Guide May 2013
The table below is taken directly from The Cochrane Handbook for Systematic Reviews of
Interventions, Section 10.1.
While there is empirical evidence that certain types of bias can substantially alter study results,
the effects of other biases on results are not yet clear. This complicates the assessment of
study quality, as ideally we only want to evaluate those sources of bias that have been shown
empirically to affect results. While the effects of some potential sources of bias on study results
are not yet clear, it is logical to suspect that certain biases, particularly selection bias,
performance bias, attrition and detection biases, may influence study results. Hence, it is
important to systematically and transparently assess studies against criteria that evaluate the
likely impact of each of these biases.
For more information, see the Cochrane Handbook for Systematic Reviews of Interventions,
Sections 8 and 10
23
Study Quality Guide May 2013
Many checklists and scoring systems have been developed and used to assist people in
assessing study quality, in particular for assessing the quality of randomised controlled trials
(RCTs). The Cochrane Collaboration has developed a tool to assist review authors to assess the
quality of studies included in reviews, known as the Risk of Bias tool. This tool specifies a
number of different domains which may predispose studies to bias, and provides a detailed
framework for assessing and reporting the risk of bias of included studies on each of these
domains.
The Risk of Bias (RoB) tool can be used to assess the quality of RCTs. It can also be adapted to
assess the quality of quasi-RCTs controlled before-and-after (CBA) studies, and interrupted
time series studies (ITS), all of which are occasionally eligible for inclusion in systematic
reviews of complex interventions such as those published through the CC&CRG.
In the Consumers and Communication Review Group, we require that the Collaboration’s Risk
of Bias tool be used. This assessment should be reported explicitly in the review in Risk of Bias
tables and review text.
Review authors should include the following text (tailored as necessary) in the
Methods section (“Assessment of Risk of Bias in Included Studies”) of their
Protocol/Review:
‘We will assess and report on the methodological risk of bias of included studies in
accordance with the Cochrane Handbook (Higgins 2011) and the guidelines of the
Cochrane Consumers and Communication Review Group (Ryan 2013), which
recommends the explicit reporting of the following individual elements for RCTs:
random sequence generation; allocation sequence concealment; blinding (participants,
24
Study Quality Guide May 2013
Studies will be deemed to be at the highest risk of bias if they are scored as at high or
unclear risk of bias for either the sequence generation or allocation concealment
domains [or other domains of the tool; please adapt], based on growing empirical
evidence that these factors are particularly important potential sources of bias (Higgins
2011).*
In all cases, two authors will independently assess the risk of bias of included studies,
with any disagreements resolved by discussion to reach consensus. We will contact
study authors for additional information about the included studies, or for clarification of
the study methods as required. We will incorporate the results of the risk of bias
assessment into the review through standard tables, and systematic narrative
description and commentary about each of the elements, leading to an overall
assessment the risk of bias of included studies and a judgment about the internal
validity of the review’s results.’
‘We will assess and report quasi-RCTs as being at a high risk of bias on the random
sequence generation item of the risk of bias tool.
‘For cluster-RCTs we will also assess and report the risk of bias associated with an
additional domain: selective recruitment of cluster participants.’
‘We will assess CBA studies against the same criteria as RCTs but report them as being
at high risk of bias on both the random sequence generation and allocation sequence
concealment items. We will exclude CBA studies that are not reasonably comparable at
baseline.’
25
Study Quality Guide May 2013
‘We will assess and report on the following items for ITS studies: intervention
independence of other changes; pre-specification of the shape of the intervention
effect; likelihood of intervention affecting data collection; blinding (participants,
personnel); blinding (outcome assessment); completeness of outcome data, selective
outcome reporting; and other sources of bias [please add].’
If you are planning to assess specific items under ‘other’ sources of bias domain this should
also be described. Note that assessing other sources of bias is not essential but should be
guided by the specific study designs that you plan to include in the review. This might
include design-specific issues (such as assessing selective recruitment of cluster
participants for cluster-RCTs), baseline imbalances between groups or the likelihood of
contamination.
26
Study Quality Guide May 2013
RCTs are considered to represent the ‘gold standard’ study design for investigating the effects
of interventions. This is because their design helps to avoid or to minimise many different
kinds of bias20 (see Study Design Guide for more details). However, it is important to
remember that not all RCTs are well-designed or well-conducted. As a result, it is possible to
have RCTs of variable quality. This in turn affects the strength of the evidence that can be
drawn from their findings (see Section 2.2.2: Study design versus study quality). Similar
principles hold for the limited range of non-randomised studies that are eligible for inclusion in
Cochrane reviews.
The Cochrane Handbook provides detailed advice, description of and practical instructions for
review authors about the Risk of Bias tool.
This includes details about:
What each domain of the tool involves;
What issues are to be considered in making a judgement about the risk of bias in each
domain; and
How to report details from the study against the tool for each domain (practical decision
rules for using the tool).
These details are captured in the RoB tables within RevMan 5, and appear within reviews so
that readers have access to this information in a systematic way. For tips on how to enter data
into RevMan 5, see “Risk of Bias” tables in the RevMan User Guide.
The RoB tool encourages transparent reporting of bias, by asking review authors not only to
make a judgement about whether particular criteria are met (or not), but also providing
information directly form the study to support the judgement made. This assists readers of the
review in understanding what the study reported and how the review authors made their
decisions about the study’s risk of bias.
Authors are advised to consult the Handbook, especially Section 8 ‘Assessing risk of bias in
included studies’ for details on the tool and how to complete the risk of bias assessments in
20
Note that bias is taken here to mean the various different sources of bias overed in earlier sections of this guide.
27
Study Quality Guide May 2013
reviews. The following sections outline the RoB tool for use by review authors with the
CC&CRG, for RCTs and quasi-RCTs, and for non-randomised studies (CBA and ITS studies).
For assessing the risk of bias of RCTs included in reviews, authors should use the Cochrane
Handbook and follow the guidance outlined.
Adapted from Cochrane Handbook Table 8.5.a: The Cochrane Collaboration’s tool for
assessing risk of bias
Domain Review authors’ Support for judgement
judgement
Describe the method used to generate the
High risk allocation sequence in sufficient detail to
Random sequence allow an assessment of whether it should
Unclear
generation* produce comparable groups.
Low risk
Quasi-RCTs must be rated as ‘High Risk’ for
random sequence generation as the methods
were not, by definition, truly random.
Describe the method used to conceal the
High risk allocation sequence in sufficient detail to
Allocation concealment Unclear determine whether intervention allocations
Low risk could have been foreseen in advance of, or
during, enrolment.
28
Study Quality Guide May 2013
Cluster RCTs are at particular risk of bias in terms of participant recruitment. We recommend
that review authors assess whether selective recruitment of cluster members was adequently
prevented during the study. This should be reported in the ‘Other Sources of Bias’ row in the
standard Risk of Bias table for RCTs.
Was selective recruitment of cluster members adequately prevented during the study?
Criteria for a judgement of Those involved in the identification and/or recruitment of the cluster
‘Low risk’ of bias. participants did not have knowledge of the group allocation because one of the
following, or an equivalent method, was employed:
Cluster participants were recruited prior to group assignment and the
same participants were followed up over time.
Cluster participants were recruited post group assignment but
o carried out by a person who was blinded to the group
allocation;
o carried out by a person who was unaware of characteristics of
the participants (e.g. clinical characteristics of patients);
o eligibility criteria were such that it was unlikely to be subverted
by knowledge of group allocation (e.g. all patients attending a
hospital within a specified period of time included);
o invited by mail questionnaire with identical information to
participants in the intervention and control arms.
21
Modified from Table 1 of Brennan et al Continuous quality improvement: effects on professional practice and
healthcare outcomes. Cochrane Database of Systematic Reviews 2009, Issue 4. Art. No.: CD003319. DOI:
10.1002/14651858.CD003319.pub2.
29
Study Quality Guide May 2013
Criteria for the judgement Those involved in the identification and/or recruitment of the cluster
of ‘High risk’ of bias. participants may have had knowledge of the group allocation.
Cluster participant identification and/or recruitment undertaken post
group allocation by a person who was unblinded and who may have
had knowledge of characteristics of the cluster participants.
Criteria for the judgement Insufficient information to permit judgement of ‘Low risk’ or ‘High risk’.
of ‘unclear risk’ of bias.
The purpose of this domain to assess the risk of selection bias from recruitment of cluster
participants into the study. The risk of selection bias is also assessed through the domains
‘Sequence generation’ and ‘Allocation concealment’ (See Chapter 8, Cochrane Handbook for
Systematic Reviews of Interventions). In cluster-randomised trials the hierarchical nature of
the design allows for differential recruitment or consent of participants, through recruitment
post-randomisation (Giraudeau B et al. PLoS Med 2009, 6:e1000065). This may potentially
introduce bias from selective recruitment irrespective of adequate sequence generation and
allocation concealment.
Selective recruitment can occur at multiple levels when there are multiple levels of clustering.
For example, in a cluster-randomised trial where primary care practices are randomised to the
intervention and control groups, selection bias could occur when recruiting practitioners to the
trial, patients, or both. Practitioners’ knowledge of their practice allocation could affect the type
of practitioners who choose to be involved in the trial and if, for example, the practitioners are
involved in the recruitment of patients, they may selectively recruit patients depending on
their clinical characteristics.
In cluster-randomised trials with multiple levels of clustering (e.g. practices, practitioners, and
patients), the risk of bias arising from selective recruitment may differ depending on the level
at which the outcome is measured. Using the above example, if all practitioners within a
practice were recruited, but practitioners recruited patients, then the risk of bias arising from
selective recruitment will be low for outcomes measured at the practitioner level, but may be
high for outcomes measured at the patient level.
The criteria in this table for judging the risk of selection bias from recruitment of cluster
participants can be applied to other designs involving clustering such as controlled clinical trials
and controlled before and after studies.
Studies of different design have different issues associated with their design and conduct.
Criteria used to evaluate the quality of RCTs cannot be applied directly (without adaptation) to
all studies of other designs that might be included in a review. However, major elements of the
30
Study Quality Guide May 2013
RoB criteria used to assess RCTs can be adapted to systematically assess non-randomised
studies.
Because controlled studies that are not randomised are similar in their design in many respects
to RCTs, they can be assessed using a similar but slightly adapted set of criteria. Assessment
of the randomisation method and suitability is obviously not relevant for a non-randomised
study. However, as allocation to intervention and control groups is not determined by
randomisation, review authors should pay particular attention to how the groups are chosen,
the possibility of bias arising from the non-random allocation of participants, and the potential
influence of baseline differences in groups upon the outcomes of the study.
At the CC&CRG we follow the guidance developed by the EPOC group, which has tailored the
RoB tool for use on study designs other than RCTs. Authors should follow the decision rules
laid out in this additional guidance, and which complements the Handbook guidance, in those
cases where studies other than RCTs are included. This guidance is presented below.
Inclusion criteria:
The EPOC Review Group specifies that, to be included in a systematic review, the CBA study
design must meet three key criteria. These are:
There must be at least two intervention sites and two control sites (note, this is a
new criterion added in 2009).
The timing of the periods for study for the control and intervention groups should be
comparable (that is, the pre- and post- intervention periods of measurement for the
control and intervention groups should be the same).
The intervention and control groups should be comparable on key characteristics.
CBA studies that do not meet these criteria should be excluded from a Cochrane
review.
31
Study Quality Guide May 2013
If CBA studies meet these criteria, they are eligible (at least based on study design criteria) for
inclusion in a systematic review and so need to be systematically assessed for their risk of
bias.
The standard Risk of Bias tool for RCTs should be utilised for this purpose, and a form suitable
for use on CBAs is shown in the table.
Adapted from Cochrane Handbook Table 8.5.a: The Cochrane Collaboration’s tool for
assessing risk of bias; adapted using EPOC’s criteria for studies other than RCTs
Other sources of bias* Note: all answers State any important concerns about bias not
For example: should follow the addressed in the other domains in the tool.
format: If particular questions/entries were pre-specified in
- Were the intervention and
32
Study Quality Guide May 2013
control groups comparable at High risk the review’s protocol, responses should be
baseline (note, if groups were Unclear provided for each question/entry.
not reasonably equivalent and
this was not adjusted through Low risk
analysis, the study should be
excluded).
- Have measures been taken
within the study to protect
against contamination?
See the Cochrane Handbook
8.15.1 for further examples of
potential threats to validity.
* If some primary outcomes were imbalanced at baseline, assessed blindly or affected by missing data
and others were not, each primary outcome can be scored separately.
Action:
Rate included CBA studies on each of the risk of bias assessment criteria. Each criteria
should be rated as low risk (done), high risk (not done), or unclear, with a description
given.
Authors should rate Random sequence generation as ‘high risk’ and Allocation
concealment as ‘high risk’ (as indicated in the table above).
Authors should pay particular attention to the element relating to baseline comparability
of intervention and control groups under the Other sources of bias heading, and
there is a high risk of groups being unbalanced in CBA designs.
Authors are referred to the guidance provided by EPOC on ITS risk of bias
assessment. Authors may wish to seek advice from the Consumers and
Communication Group (cochrane@latrobe.edu.au) on adapting this to their reviews.
The EPOC Review Group specifies that, to be included in a systematic review, studies of the
ITS design must meet two key criteria. These are
33
Study Quality Guide May 2013
There must be a clearly defined point in time at which the intervention occurred,
and this should be reported by the researchers.
There should be collection of at least three data points before and three after the
intervention was introduced.
ITS studies that do not meet these criteria should be excluded from a Cochrane
review.
If ITS studies meet these criteria, they are eligible (at least on study design criteria) for
inclusion in a systematic review and so need to be systematically assessed for their risk of
bias.
The EPOC guidelines for risk of bias assessment of ITS studies are available on their website.
Note that the scope of EPOC focuses more on healthcare structures and organisational aspects
than does the scope of the Consumers and Communication Review Group. Review authors
should bear this in mind, and will need to adapt the EPOC guidance to suit the needs of their
specific review question.
Adapted from Cochrane Handbook Table 8.5.a: The Cochrane Collaboration’s tool for
assessing risk of bias; adapted using EPOC’s criteria for ITS studies and CCRG input
Was the intervention High Risk Score “Low Risk” if reported that
unlikely to affect data intervention itself was unlikely to affect
34
Study Quality Guide May 2013
35
Study Quality Guide May 2013
*** If some primary outcomes were assessed blindly or affected by missing data and others were not,
each primary outcome can be scored separately.
36
Study Quality Guide May 2013
NHS Centre for Reviews and Dissemination (2009) Systematic reviews: CRD’s guidance for
undertaking reviews in health care, available at
http://www.york.ac.uk/inst/crd/index_guidance.htm, Chapter 6: Incorporating qualitative
evidence in or alongside effectiveness reviews.
Sometimes it is both appropriate and useful to include qualitative data as part of a systematic
review. It is now becoming more common to include data from qualitative research alongside
quantitative data in systematic reviews of effectiveness. Qualitative data can, for example,
help to provide depth and explanation for observed outcomes; help to decide whether the
intervention of interest is suitable for a particular target population; help to examine factors
that might have influenced the results if the effect of the intervention is not in the expected
direction; and so on. Such contributions can add meaning to the results of a systematic review
and can help to address some of the questions which are likely to be important to the users of
systematic reviews (for details, see Study Design Guide, Quantitative and qualitative
designs).
Note that inclusion of qualitative data in the analysis section of a review must be
clearly justified by the authors and is subject to editorial approval.
22
For example, the Critical Appraisal Skills Program (CASP) tool for qualitative studies.
37
Study Quality Guide May 2013
from included studies). Despite these differences, there is agreement that the aim of quality
assessment of qualitative research should be to adopt a structured and transparent approach.
The Cochrane Qualitative Research Methods Group and other groups worldwide are in the
process of developing standards for the assessment and reporting of qualitative data.
Guidelines and recommendations for review authors will be developed and this document
updated when they become available.
Authors must contact the editorial base of the Consumers and Communication Group
directly if they wish to incorporate qualitative data into their review.
38
Study Quality Guide May 2013
A structured and explicit approach to assessing and reporting the risk of bias of included
studies is an important component of a systematic review. While this information must be
included in the review (for example, as a component of the ‘Table of Included Studies’ or as a
separate table or figure generated by RevMan), this information is not all that is needed.
The text of the systematic review should attempt to integrate the risk of bias assessment by
using these assessments to interpret the findings of the studies, and to draw conclusions.
In terms of internal validity, as well as discussing the individual risk of bias items considered
for different studies, you might attempt to draw this together by considering some of the
following, or similar, types of questions:
How confident are you that the results of the included studies reflect what is true?
Are there limitations in the design or execution of the studies that might render the
results less believable? How severe are these limitations?
Were there aspects of the studies that were particularly well designed and so increase
your confidence in the results?
In terms of the external validity or generalisability of the studies, you might wish to consider
and discuss some of the following types of questions when considering the body of evidence
you have gathered and appraised as a whole:
39
Study Quality Guide May 2013
What works for whom? Which groups of the population do the results apply to? (e.g.
Was the study only of children? Of adults? Of people with a particular disease but no
comorbidities?)
Has the intervention been evaluated in a particular setting only? (e.g. Was the
intervention delivered in a hospital setting only? Are the results likely to be applicable
to the community setting?)
Do some versions of the intervention work better or best? (e.g. Is the intensity or
frequency of delivery of the intervention likely to affect results? Are some versions
ineffective? Are some versions effective only in some people?)
Narratively describe and comment upon the risk of bias issues associated with the
included studies for each of the rated items. Some issues to consider when trying to
formulate an overall description of the body of evidence gathered in the review might
include the following:
o Did most of the included trials report adequate detail on their methods of sequence
generation (or other allocation methods)?
o Have you had to assume that they used adequate methods due to lack of
information?
o Did many studies adequately conceal allocation, and what methods did they use?
How do the results of these studies compare to those that did not adequately
conceal allocation?
o Was blinding done adequately (for participants, providers, outcome assessors and
analysts; for all outcome measures)?; or was it not possible to blind all of those
involved in the study?
o Was follow-up of participants throughout the study period described; and was it
adequate?
Draw some general comments on the overall risk of bias of studies included in the
review. The purpose of doing so is to help to set the scene for the Results and Discussion
sections of the review. As the results of the studies need to be interpreted in the light of
the risk of bias of the studies, this is a critical part of your review. Some of the things you
may wish to consider might include:
o Are there particular limitations in the design, execution or reporting of the included
studies that render the results less believable?
o How severe are these limitations?
40
Study Quality Guide May 2013
o Were there aspects of the included studies that were particularly well designed and
so increase your confidence in the results?
It may be appropriate at this point, for example, to highlight any particular strengths
and/or limitations of the included studies. The main focus should be to discuss and attempt
to narratively synthesise the issues associated with the risk of bias of the included studies.
For reviews where studies other than RCTs have also been included, we suggest that
studies of different designs be discussed as separate sections in terms of their risk of
bias and each of their respective strengths and weaknesses.
o For example, it might be appropriate to discuss the risk of bias of RCTs and non-
randomised controlled studies separately: as their designs are different, the
discussion will need to deal with different aspects relating to design. Once this is
done, it may then be appropriate to contrast the different types of studies, and to
draw comparisons between the different categories.
o It can be especially interesting, for example, to compare the findings of studies of
different design and varying levels of quality to comment on the consistency of the
findings of studies included in the review. For example, do non-randomised studies
report the same direction and size of results as RCTs? Do studies with a low risk of
bias report the same direction and size of results as studies with a high risk of bias?
41
Study Quality Guide May 2013
ATTRITION BIAS
Tierney JF. and Stewart LA. Theory and Methods: Investigating patient exclusion bias in meta-
analysis. International Journal of Epidemiology 2005; 34 (1): 79-87.
Juni P. and Egger M. Commentary: Empirical evidence of attrition bias in clinical trials.
International Journal of Epidemiology 2005; 34 (1): 87-8.
This study (and commentary) analyses the effects of participant exclusion from trials (post-
randomisation), and directly analyses the effects of attrition bias upon trial effect estimates.
It includes a useful glossary of terms.
This study demonstrates that pooled analysis of trials that excluded patients post-
randomisation may be prone to bias. The authors conclude that analysis (in trials, meta-
analyses and systematic reviews) should be based on all participants randomised at the start
of the trial.
BLINDING
Devereaux PJ, Manns BJ, Ghali WA, Quan H, Lacchetti C, Montori VM, Bhandari M and Guyatt GH.
Physician interpretations and textbook definitions of blinding terminology in randomised
controlled trials. JAMA 2001; 285(15): 2000-5.
As physicians commonly use study blinding as a quality assessment criteria for studies they
read, this study examined the definitions and interpretations of blinding used by physicians
and by textbooks.
42
Study Quality Guide May 2013
The study reports little consensus about the use of blinding and associated terms (single,
double, etc); and discusses the findings in terms of implications for the quality of trials.
Spiegelhalter DJ, Myles JP, Jones DR, Abrams KR. (2000) Bayesian methods in health technology
assessment: a review. Health Technol Assess 4(34).
http://www.hta.ac.uk/execsumm/summ438.htm
This paper assesses the association between study (methodological) quality and estimates of
effect sizes. In particular, it examines the differences between effect sizes generated by RCTs
in comparison with quasi-randomised and observational studies. It discusses differences in
effect estimates across various different non-randomised study designs.
43
Study Quality Guide May 2013
This study also recommends that further studies to directly compare the results of RCTs and
quasi-randomised and observational studies be performed; and that further instruments be
developed to assess study quality and the accuracy of effect estimates in relation to study
quality.
PUBLICATION BIAS
Song F, Eastwood AJ, Gilbody S, Duley L, Sutton AJ. Publication and related biases. Health
Technol Assess 2000;4(10) http://www.hta.ac.uk/execsumm/summ410.shtml
This paper systematically examines and evaluates studies on methodological issues
associated with publication bias. It attempts to identify empirical evidence about the
existence and consequences of publication bias and other forms of dissemination-based
biases.
For further discussion of issues associated with public health interventions, also see the
Cochrane Health Promotion and Public Health Review Group. Guidelines for Systematic Reviews
of Health Promotion and Public Health Interventions. http://ph.cochrane.org/review-authors
44
Study Quality Guide May 2013
QUALITY ASSESSMENT
Jüni P, Altman DG and Egger M. Assessing the quality of controlled clinical trials. BMJ 2001; 323:
42-6.
This paper discusses the various different elements of quality that are necessary to assess
for studies included in systematic reviews.
It provides a detailed description of different sources of bias in studies of healthcare
interventions; as well as a description of the ways in which study quality can be incorporated
into meta-analysis and systematic reviews.
Moher D, Cook DJ, Jadad AR, Tugwell P, Moher M, Jones A, et al. Assessing the quality of reports
of randomised trials: implications for the conduct of meta-analyses. Health Technol Assessment
1999; 3(12) http://www.hta.ac.uk/execsumm/summ312.shtml
This paper examines and discusses the various different elements of quality that are
necessary to assess for studies included in systematic reviews.
It discusses in detail the different aspects of study quality; which are likely to be most
important (ie. to impact the most upon results), and makes recommendations for assessing
study quality as part of systematic review conduct.
Schultz KF, Chalmers I, Hayes RJ and Altman DG. Empirical evidence of bias: Dimensions of
methodological quality associated with estimates of treatment effects in controlled trials. JAMA
1995; 273(5): 408-12.
This review assessed the relationships between different aspects of trial quality and trial
results.
It reports empirical evidence that several elements of study quality (allocation concealment
and blinding), if not adequately performed, tend to overestimate the estimates of effect in
trials.
Centre for Reviews and Dissemination. Systematic reviews: CRD's guidance for undertaking
reviews in health care. York: University of York; 2009 [Available from:
http://www.york.ac.uk/inst/crd/pdf/Systematic_Reviews.pdf] see Chapter 1.3.4 in particular.
These resources give an overview of study quality issues, bias and why it is necessary to
assess the quality of studies included in systematic review. It also provides ways of
approaching quality assessment of studies, with discussion in terms of answering different
types of research questions.
45
Study Quality Guide May 2013
SEARCHING
Egger M, Juni P, Bartlett C, Holenstein F, Sterne J. (2003) How important are comprehensive
literature searchers and the assessment of trial quality in systematic reviews: empirical study.
HTA Methodology 7(26). http://www.hta.ac.uk/pdfexecs/summ701.pdf
This paper examines the characteristics of trials that are difficult to locate by searching. It
also examines the effects of including only easily accessible studies, compared with more
difficult to locate studies (that require comprehensive search techniques), on reported
treatment effects and on pooled treatment effects (from meta-analysis).
It concludes that trials that are difficult to locate tend to be smaller and of poorer
methodological quality than those that are easily accessible; and that the inclusion or
exclusion of trials of low methodological quality has a substantial impact on the results and
the conclusions reached by systematic reviews and meta-analyses.
See also Jadad 1998.
CONSORT STATEMENT
Moher D., Schulz K.F. and Altman D.G. The CONSORT statement: revised recommendations for
improving the quality of report of parallel-group randomised trials. The Lancet 2001; 357 (9263):
1191-8.
This paper outlines the CONSORT checklist, which has been developed to improve the
reporting of RCTs, so that readers will be better able to determine why the study was
undertaken, and how it was conducted and analysed.
A copy of the most recent version of the CONSORT checklist (2010) is available at
http://www.consort-statement.org/index.aspx?o=1031
TREND STATEMENT
Des Jarlais D.C., Lyles C., Crepaz N. and the TREND Group. Improving the reporting quality of
nonrandomised evaluations of behavioural and public health interventions: The TREND
statement. American Journal of Public Health 2004; 94(3): 361-6.
This paper notes that non randomised trials and other studies should be included in
systematic reviews, as excluding all data from studies other than RCTs will bias the evidence
base towards those interventions which are ‘simpler’ to evaluate (ie. they can be evaluated
using an RCT). This forms the basis for the development of the TREND statement, a
consensus statement for the reporting of nonrandomised intervention studies.
The authors describe the TREND checklist and its development (as companion to the
CONSORT statement for RCTs), which provides guidelines for the transparent reporting of
studies, including aspects of intervention and comparisons, the research design, as well as
methods of adjusting for possible sources of bias.
A copy of the TREND checklist is available at http://www.cdc.gov/trendstatement
46
Study Quality Guide May 2013
REFERENCES
Cochrane Collaboration Open Learning Material for Reviewers, Version 1.1, November 2002.
Eds, P Alderson and S Green, www.cochrane-net.org/openlearning/index.htm
Cochrane Health Promotion and Public Health Field. Guidelines for Systematic Reviews of
Health Promotion and Public Health Interventions. http://ph.cochrane.org/review-authors
Cochrane Effective Practice and Organisation of Care (EPOC) Review Group. The data collection
checklist. http://epoc.cochrane.org/epoc-resources
Cochrane Effective Practice and Organisation of Care (EPOC) Review Group. The data collection
template. http://epoc.cochrane.org/epoc-resources
Cochrane Effective Practice and Organisation of Care (EPOC) Review Group. EPOC Methods
Paper: Including Interrupted Times Series (ITS) designs in an EPOC review.
http://epoc.cochrane.org/epoc-resources
Cochrane Effective Practice and Organisation of Care (EPOC) Review Group. EPOC Methods
Paper: Issues Related to Baseline Measures of Performance. http://epoc.cochrane.org/epoc-
resources
Systematic reviews: CRD's guidance for undertaking reviews in health care. Centre for Reviews
and Dissemination. York: University of York; 2009 [Available from:
http://www.york.ac.uk/inst/crd/pdf/Systematic_Reviews.pdf]
Deeks JJ, Dinnes J, D’Amico R, Sowden AJ, Sakarovitch C, Song F, et al. (2003) Evaluating non-
randomised intervention studies. Health Technol Assess 7(27).
http://www.hta.ac.uk/execsumm/summ727.htm
Egger M, Jüni P, Bartlett C, Holenstein F, Sterne J. (1999) How important are comprehensive
literature searches and the assessment of trial quality in systematic reviews? Empirical
study. Health Technol Assess 7(1). http://www.hta.ac.uk/execsumm/summ701.htm
Heritier SR, Gebski VJ and Keeck AC (2003). Inclusion of patients in clinical trials analysis: the
intention-to-treat principle. MJA, 179: 438-40.
Higgins JPT, Green S (editors). Cochrane Handbook for Systematic Reviews of Interventions
Version 5.1.0 [updated March 2011]. The Cochrane Collaboration, 2011. Available from
www.cochrane-handbook.org.
Jadad A. (1998) ‘Randomised controlled trials: A user’s guide.’ BMJ Books, London.
Juni P, Witschi A, Bloch R, Egger M. (1999) The hazards of scoring the quality of clinical trials
for meta-analysis. JAMA; 282: 1054-60.
Moher D, Cook DJ, Jadad AR, Tugwell P, Moher M, Jones A, et al. (1999) Assessing the quality
of reports of randomised trials: implications for the conduct of meta-analyses. Health
Technol Assessment 3(12).
Song F, Eastwood AJ, Gilbody S, Duley L, Sutton AJ. Publication and related biases. Health
Technol Assess 2000;4(10)
http://www.york.ac.uk/inst/crd/projects/publication_bias.htm
47
Study Quality Guide May 2013
Spiegelhalter DJ, Myles JP, Jones DR, Abrams KR. (2000) Bayesian methods in health
technology assessment: a review. Health Technol Assess 4(34).
http://www.hta.ac.uk/execsumm/summ438.htm
Torgerson DJ. and Sibbald B. What is a patient preference trial? BMJ 1998; 316: 360.
48